Bayesian Methods for Meta-Analysis with Applications to Multi Arm Trials with Binary Outcomes

Recently, there has been a growing interest in meta-analysis in many areas including medicine, education, psychology, social sciences. In literature, there are two main approaches used in meta-analysis: the fixed effect model and the random effects model. Under the fixed effect model we assume that there is one true effect size which is shared by all the included studies. The combined effect is our estimate of this common effect size. If there is no statistical heterogeneity among studies, differences across studies may be due to random variation and fixed effects model may be appropriate. In random effects model, we assume that the true effect could vary from study to study. Muthukumarana and Tiwari (2012) developed a random-effects model using Dirichlet process priors to account for heterogeneity among studies. This project will focus on enhancing the methodology developed in this paper. More specifically, the methodology will be extended to multivariate version of random meta-analysis with binary outcomes. This extension is important when there are several zeros observed in some studies. This will compel to introduce zero-inflated Binomial (ZIB) models for meta-analysis.

Faculty Supervisor: 
Saman Muthukumarana
Department: 
Statistics
Partner University: 
University of Manitoba